A Predictive Model for the Identification of Cardiac Effusions Misclassified by Light’s Criteria

Abstract

Abstract Objectives The application of Light’s criteria misidentifies approximately 30% of transudates as exudates, particularly in patients on diuretics with cardiac effusions. The purpose of this study was to establish a predictive model to effectively identify cardiac effusions misclassified by Light’s criteria. Methods We retrospectively studied 675 consecutive patients with pleural effusion diagnosed by Light’s criteria as exudates, of which 43 were heart failure patients. A multivariate logistic model was developed to predict cardiac effusions. The performance of the predictive model was assessed by receiver operating characteristic (ROC) curves, as well as by examining the calibration. Results It was found that protein gradient of >23 g/L, pleural fluid lactate dehydrogenase (PF-LDH) levels, ratio of pleural fluid LDH to serum LDH level (P/S LDH), pleural fluid adenosine deaminase (PF-ADA) levels, and N-terminal pro–brain natriuretic peptide (NT-pro-BNP) levels had a significant impact on the identification of cardiac effusions, and those were simultaneously analyzed by multivariate regression analysis. The area under the curve (AUC) value of the model was 0.953. The model also had higher discriminatory properties than protein gradients (AUC, 0.760) and NT-pro-BNP (AUC, 0.906), all at a P value of <.01. Conclusion In cases of suspected cardiac effusion, or where clinicians cannot identify the cause of an exudative effusion, this model may assist in the correct identification of exudative effusions as cardiac effusions.